Back to Details

review-code

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Skill: Review Code (Orchestrator)

Skill：代码审查（编排器）

Purpose

用途

This skill does not perform code analysis itself. It is a meta skill that orchestrates other atomic review skills in a fixed order, then aggregates their findings into a single report. Use it when the user asks for a full "review code" or "code review" and you want to apply scope → language → framework → library → cognitive skills and produce one combined output. For a single-dimension review (e.g. only diff or only security), invoke the corresponding atomic skill directly (review-diff, review-security, etc.).
该Skill自身不执行代码分析。它是一款元Skill，按固定顺序编排其他原子审查Skill，然后将所有审查结果汇总成一份单一报告。当用户要求进行完整的“代码审查”，且你希望按范围→语言→框架→库→认知类Skill的流程生成一份合并输出时，可使用本Skill。如果是单维度审查（如仅审查diff或仅审查安全），请直接调用对应的原子Skill（review-diff、review-security等）。

Use Cases

使用场景

  • Full code review: User asks to "review code" or "review my changes" and expects impact, language/framework conventions, security, and architecture in one pass.
  • Pre-PR or pre-commit: Run the full pipeline (diff + stack + cognitive) and get one report.
  • Consistent pipeline: Same execution order every time so Cursor or an agent can simulate skill chaining by following this skill's instructions.
When to use: When the user wants a combined review across scope, stack (language/framework/library), and cognitive dimensions. When the user wants only one dimension (e.g. "review my diff" or "security review"), use the atomic skill instead.
  • 完整代码审查：用户要求“审查代码”或“审查我的修改”，期望一次性获取影响评估、语言/框架规范、安全及架构相关结果。
  • PR前或提交前检查：运行完整流水线（diff + 技术栈 + 认知类审查）并生成一份报告。
  • 一致的流水线：每次执行顺序固定，Cursor或Agent可遵循本Skill的说明模拟Skill链式调用。
适用时机：当用户需要跨范围、技术栈（语言/框架/库）和认知维度的合并审查时。若用户仅需单维度审查（如“审查我的diff”或“安全审查”），请使用对应的原子Skill。

Behavior

行为规范

Orchestration only

仅负责编排

  • Do not analyze code yourself. Do invoke (or simulate invoking) the following skills in order, then aggregate their findings.
  • Execution order is fixed so that Cursor or an agent can follow it step by step.
  • 不要自行分析代码。务必按顺序调用（或模拟调用）以下Skill，然后汇总它们的审查结果。
  • 执行顺序固定，确保Cursor或Agent可逐步遵循。

Interaction policy

交互规则

  • Prefer defaults and choices: Use the defaults in the table below; present options for the user to confirm or select (e.g. [diff] [codebase], [Repo root] [Current dir]), and avoid asking for free-text input when a default exists.
  • Scope (diff vs codebase): If the user has not explicitly indicated (a) diff/current change (e.g. "my changes", "the diff", "what I changed") or (b) codebase/path (e.g. "this directory", "src/foo", "the repo"), ask the user to choose. In particular, if they said only "review", "review code", or "code review" with no scope cue, do not assume — offer: Review current change (diff) [default] or codebase (given path(s))? and wait for their choice before running any review skill.
  • If language/framework is not explicit and cannot be inferred from the files in scope, offer choices ([.NET] [Java] [SQL] [Vue] [Skip]); if still unclear, skip and note the skip in the final summary.
  • Always state which steps were executed and which were skipped (with reason).
  • 优先使用默认值和选项选择：使用下表中的默认值；为用户提供确认或选择的选项（如[diff] [代码库]、[仓库根目录] [当前目录]），当存在默认值时避免要求用户输入自由文本。
  • 范围（diff vs 代码库）：若用户未明确指定（a）diff/当前修改（如“我的修改”“diff”“我改的内容”）或（b）代码库/路径（如“此目录”“src/foo”“仓库”），请让用户选择。特别是当用户仅说“审查”“审查代码”或“代码审查”而未提及范围时，不要自行假设，请提供选项：审查当前修改（diff） [默认] 还是代码库（指定路径）？，等待用户选择后再运行任何审查Skill。
  • 若无法从范围中的文件推断出语言/框架，提供选项（[.NET] [Java] [SQL] [Vue] [跳过]）；若仍不明确，则跳过该步骤并在最终摘要中注明跳过原因。
  • 始终说明已执行的步骤和跳过的步骤（及原因）。

Defaults (prefer confirm or choose; avoid asking for free-text input)

默认值（优先确认或选择；避免要求自由文本输入）

ItemDefaultWhen to deviate
Scopediff (current change)User chooses "codebase" to review given path(s) instead.
Scope = diff — untrackedInclude untracked files in change setUser can choose "diff only, no untracked."
Scope = codebase — path(s)Repo rootUser chooses one or more paths (offer: repo root / current file’s dir / list top-level dirs to pick).
Scope = codebase — largeBy layer (output by module/dir; no single shallow pass)User can choose a priority subset (e.g. one layer or named modules).
Language / frameworkInfer from files in scopeIf unclear, offer choices: [.NET] [Java] [SQL] [Vue] [Skip]; do not ask user to type.
项默认值偏离时机
范围diff（当前修改）用户选择“代码库”来审查指定路径时。
范围 = diff — 未追踪文件包含变更集中的未追踪文件用户可选择“仅diff，不包含未追踪文件”。
范围 = 代码库 — 路径仓库根目录用户选择一个或多个路径（提供选项：仓库根目录 / 当前文件所在目录 / 列出顶级目录供选择）。
范围 = 代码库 — 大型代码库按层级（按模块/目录输出；不做单次浅层扫描）用户可选择优先级子集（如某一层级或指定模块）。
语言 / 框架从范围中的文件推断若无法推断，提供选项：[.NET] [Java] [SQL] [Vue] [跳过]；不要让用户手动输入。

Pre-flight: confirm before running

执行前确认

Resolve the following with the user once, before executing any review step. Prefer confirm default or select from options; avoid asking for free-text input when a default exists.
ItemIf unclearAction
ScopeUser did not say "my changes"/"diff" vs "codebase"/path (e.g. "review" or "review code" alone = unclear)Must ask. Offer: Review current change (diff) [default] or codebase (given path(s))? — user chooses; do not assume.
Scope = diff—Confirm: Include untracked files? Default Yes. Ensure diff + untracked content available for review-diff.
Scope = codebasePath(s) not statedOffer: Review repo root? [default] Or pick path(s): [repo root] [current file’s dir] [list top-level dirs] — user selects, no typing.
Scope = codebase, largeWhole repo or very large dirDefault: output by layer (module/dir). Option: Narrow to a priority subset? — user can choose from listed dirs/modules.
Language / frameworkCannot infer from filesOffer: [.NET] [Java] [SQL] [Vue] [Skip] — user picks one; if Skip or none match, skip and note in summary.
After pre-flight, run the pipeline without further scope questions; report which steps ran and which were skipped.
在执行任何审查步骤前，先与用户确认以下事项。优先使用确认默认值或选择选项**；当存在默认值时避免要求自由文本输入。**
项不明确时操作
范围用户未说明是“我的修改”/“diff”还是“代码库”/路径（如仅说“审查”或“审查代码” = 不明确）必须询问。提供选项：审查当前修改（diff） [默认] 还是代码库（指定路径）？ — 由用户选择；不要自行假设。
范围 = diff—确认：是否包含未追踪文件？ 默认是。确保diff + 未追踪内容可用于review-diff审查。
范围 = 代码库未指定路径询问：审查仓库根目录？ [默认] 还是选择路径：[仓库根目录] [当前文件所在目录] [列出顶级目录] — 由用户选择，无需输入。
范围 = 代码库，大型代码库整个仓库或超大目录默认：按层级输出（模块/目录）。选项：是否缩小到优先级子集？ — 用户可从列出的目录/模块中选择。
语言 / 框架无法从文件推断提供选项：[.NET] [Java] [SQL] [Vue] [跳过] — 用户选择其一；若选择跳过或无匹配项，则跳过该步骤并在摘要中注明。
完成执行前确认后，无需再询问范围相关问题，直接运行流水线；报告中需说明已执行的步骤和跳过的步骤。

Execution order

执行顺序

When performing this skill, sequentially apply the following steps. For each step, load and run the corresponding skill's instructions, collect its findings (in the standard format: Location, Category, Severity, Title, Description, Suggestion), then proceed to the next step.
  1. Scope
    Choose one based on user intent:
    • review-diff: Use when the user wants only the current change (git diff, staged + unstaged) reviewed. Load review-diff and run it on the diff.
    • review-codebase: Use when the user wants the current state of given path(s), directory(ies), or repo reviewed. Load review-codebase and run it on the specified scope. Run the chosen scope skill; collect all findings.
  2. Language
    Choose one or none based on the project's primary language in scope:
    • review-dotnet: .NET (C#/F#). Load review-dotnet.
    • review-java: Java. Load review-java.
    • review-sql: SQL or query-heavy code. Load review-sql. If none match, skip this step. Run the chosen language skill on the same scope; collect all findings.
  3. Framework (optional)
    If the project uses a known framework in scope, choose the matching skill:
    • review-vue: Vue 3. Load review-vue.
    • (Reserved for future: review-aspnetcore, review-react, etc.) If none match, skip. Run the chosen framework skill; collect all findings.
  4. Library (optional)
    If the project heavily uses a key library with a dedicated review skill, run it (e.g. review-entityframework when available). Otherwise skip. Collect all findings.
  5. Cognitive
    Run in order:
    • review-security: security findings.
    • review-architecture: architecture findings. (Reserved for future: review-reliability, review-maintainability.)
      Collect all findings.
  6. Aggregation
    Merge all collected findings into one report. Group by Category (
    scope
    , 
    language-*
    , 
    framework-*
    , 
    library-*
    , 
    cognitive-*
    ) or by file/location, as best fits the report length. Use the same finding format (Location, Category, Severity, Title, Description, Suggestion). Add a short summary (e.g. counts by severity or category) at the top if useful.
    De-dup rule: If multiple findings share the same Location + Title and represent the same issue across steps, keep the highest severity and note the other step(s) in the Description (e.g. "Also flagged by language and security").
执行本Skill时，请按顺序应用以下步骤。对于每个步骤，加载并运行对应Skill的说明，收集其审查结果（采用标准格式：位置、类别、严重程度、标题、描述、建议），然后进入下一步。
  1. 范围
    根据用户意图选择其一：
    • review-diff：当用户仅需审查当前修改（git diff、暂存+未暂存内容）时使用。加载review-diff并在diff上运行。
    • review-codebase：当用户需审查指定路径、目录或仓库的当前状态时使用。加载review-codebase并在指定范围上运行。 运行所选的范围Skill；收集所有结果。
  2. 语言
    根据范围内项目的主要语言选择其一或跳过：
    • review-dotnet：适用于.NET（C#/F#）。加载review-dotnet。
    • review-java：适用于Java。加载review-java。
    • review-sql：适用于SQL或查询密集型代码。加载review-sql。 若无匹配项，则跳过该步骤。在同一范围上运行所选的语言Skill；收集所有结果。
  3. 框架（可选）
    若范围内项目使用已知框架，选择匹配的Skill：
    • review-vue：适用于Vue 3。加载review-vue。
    • （预留未来扩展：review-aspnetcore、review-react等） 若无匹配项，则跳过该步骤。运行所选的框架Skill；收集所有结果。
  4. 库（可选）
    若项目大量使用有专属审查Skill的核心库，则运行该Skill（如可用的review-entityframework）。否则跳过该步骤。收集所有结果。
  5. 认知类
    按顺序运行：
    • review-security：获取安全相关结果。
    • review-architecture：获取架构相关结果。 （预留未来扩展：review-reliability、review-maintainability。） 收集所有结果。
  6. 汇总
    将所有收集到的结果合并为一份报告。根据报告长度，按类别（
    scope
    、
    language-*
    、
    framework-*
    、
    library-*
    、
    cognitive-*
    ）或文件/位置分组。采用统一的结果格式（位置、类别、严重程度、标题、描述、建议）。若有用，可在顶部添加简短摘要（如按严重程度或类别统计数量）。
    去重规则：若多个结果的位置+标题相同，且代表不同步骤中的同一问题，则保留严重程度最高的结果，并在描述中注明其他步骤（如“同时被语言审查和安全审查标记”）。

Summary for Cursor/Agent

给Cursor/Agent的摘要

  • When performing this skill, sequentially apply:
    1. review-diff or review-codebase (scope)
    2. review-dotnet or review-java or review-sql (language, optional)
    3. review-vue or other framework skill (optional)
    4. Library skill (optional, when available)
    5. review-security, then review-architecture (cognitive)
  • Aggregate all findings into a single report using the standard findings format. Do not analyze code in this skill; only orchestrate and aggregate.
  • 执行本Skill时，请按顺序应用：
    1. review-diff 或 review-codebase（范围）
    2. review-dotnet 或 review-java 或 review-sql（语言，可选）
    3. review-vue或其他框架Skill（可选）
    4. 库Skill（可选，若可用）
    5. review-security，然后是review-architecture（认知类）
  • 将所有结果汇总成一份报告，采用标准结果格式。本Skill不进行代码分析；仅负责编排和汇总。

Input & Output

输入与输出

Input

输入

  • User intent: What to review (e.g. "my changes" → scope = diff; "this directory" → scope = codebase) and optionally project type (e.g. .NET, Java, Vue) to select language/framework.
  • Code scope: Diff or paths, as provided by the user when invoking the skill.
  • 用户意图：要审查的内容（如“我的修改”→范围=diff；“此目录”→范围=代码库），以及可选的项目类型（如.NET、Java、Vue）用于选择语言/框架。
  • 代码范围：用户调用Skill时提供的diff或路径。

Output

输出

  • Single aggregated report containing all findings from the steps above, in the standard format (Location, Category, Severity, Title, Description, Suggestion), grouped by category or location, with optional summary.
  • 单一汇总报告，包含上述所有步骤的结果，采用标准格式（位置、类别、严重程度、标题、描述、建议），按类别或位置分组，可包含可选摘要。

Restrictions

限制

  • Do not perform any code analysis inside this skill. Only orchestrate other skills and aggregate.
  • Do not change the execution order; keep scope → language → framework → library → cognitive.
  • Do not invent findings; only include findings produced by the atomic skills you run.
  • 不要在本Skill内执行任何代码分析。仅负责编排其他Skill并汇总结果。
  • 不要更改执行顺序；保持范围→语言→框架→库→认知的顺序。
  • 不要编造结果；仅包含所运行原子Skill生成的结果。

Self-Check

自检清单

  • Were pre-flight items (scope, paths if codebase, large-scope priority, untracked if diff) confirmed with the user before running?
  • Was the execution order followed (scope → language → framework → library → cognitive)?
  • Were findings only collected from the atomic skills, not invented?
  • Is the output a single report with all findings in the standard format?
  • Did this skill refrain from analyzing code directly?
  • 执行前是否已与用户确认所有前置事项（范围、代码库路径、大型范围优先级、diff是否包含未追踪文件）？
  • 是否遵循了执行顺序（范围→语言→框架→库→认知）？
  • 是否仅收集原子Skill生成的结果，未编造内容？
  • 输出是否为采用标准格式的单一报告？
  • 本Skill是否未直接分析代码？

Examples

示例

Example 1: Diff review for .NET project

示例1：.NET项目的diff审查

  • Input: User says "review my code" and provides a git diff; project is C#.
  • Expected: Run review-diff → review-dotnet → review-security → review-architecture (skip framework/library if not Vue or other); aggregate all findings into one report with categories 
    scope
    , 
    language-dotnet
    , 
    cognitive-security
    , 
    cognitive-architecture
    .
  • 输入：用户说“审查我的代码”并提供git diff；项目为C#。
  • 预期结果：运行review-diff → review-dotnet → review-security → review-architecture（若不是Vue或其他框架/库则跳过）；将所有结果汇总成一份报告，包含
    scope
    、
    language-dotnet
    、
    cognitive-security
    、
    cognitive-architecture
    类别。

Example 2: Codebase review for Vue frontend

示例2：Vue前端的代码库审查

  • Input: User says "review src/frontend" and project uses Vue 3.
  • Expected: Run review-codebase on src/frontend → review-vue → review-security → review-architecture; aggregate into one report.
  • 输入：用户说“审查src/frontend”且项目使用Vue 3。
  • 预期结果：在src/frontend上运行review-codebase → review-vue → review-security → review-architecture；汇总成一份报告。

Edge case: No language match

边缘案例：无匹配语言

  • Input: Project is Rust or another language with no atomic skill yet.
  • Expected: Run scope (review-diff or review-codebase) → skip language and framework → run review-security and review-architecture; aggregate. Report should note that language/framework review was skipped (no matching skill).
  • 输入：项目为Rust或其他暂无对应原子Skill的语言。
  • 预期结果：运行范围Skill（review-diff或review-codebase）→ 跳过语言和框架步骤 → 运行review-security和review-architecture；汇总结果。报告中需注明语言/框架审查已跳过（无匹配Skill）。

Appendix: Output contract

附录：输出约定

The aggregated report MUST use the same finding format as the atomic skills:
ElementRequirement
Location
path/to/file.ext
(optional line or range).
Category
scope
, 
language-*
, 
framework-*
, 
library-*
, 
cognitive-*
Severity
critical
, 
major
, 
minor
, 
suggestion
.
TitleShort one-line summary.
Description1–3 sentences.
SuggestionConcrete fix or improvement (optional).
Group findings by Category or by Location. Optionally include a summary table (e.g. count by severity or by category) at the top of the report.
汇总报告必须采用与原子Skill一致的结果格式：
元素要求
位置
path/to/file.ext
（可选行号或范围）。
类别
scope
、
language-*
、
framework-*
、
library-*
、
cognitive-*
严重程度
critical
、
major
、
minor
、
suggestion
。
标题简短的单行摘要。
描述1-3句话。
建议具体的修复或改进方案（可选）。
结果可按类别或位置分组。可选择在报告顶部添加摘要表格（如按严重程度或类别统计数量）。